Chain-of-thought prompting
Chain-of-thought (CoT) prompting is a technique that asks a language model to work through intermediate reasoning steps before giving a final answer. Described by Wei et al. in 2022, it improves accuracy on multi-step problems such as arithmetic, logic and planning, and it inspired later reasoning models.
In the original method, the prompt includes examples whose answers show the reasoning step by step, so the model imitates that structure. A simpler zero-shot version adds an instruction such as "Let's think step by step" (Kojima et al., 2022). Writing out steps gives the model more computation per problem and makes each step build on the previous one, reducing jumps to wrong conclusions.
Why it matters: chain-of-thought helps when a task needs several dependent steps, such as calculating a quote, diagnosing an error or comparing options against criteria. Reasoning models now perform this process internally, so explicit chain-of-thought prompts matter most with standard models. Asking for visible steps also makes it easier to see where an answer went wrong.
Example: rather than asking "How much do 3 licences cost for 14 months with a 10% annual-billing discount?", ask the model to list the monthly price, the number of months, the discount and the total as separate steps before answering.
Limits: written reasoning can look convincing while containing errors, and it is not always a faithful account of how the model reached its answer. It also adds output tokens, which increases cost and latency.
Related terms
Learn it step by step
Sources
- Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. Wei et al., arXiv, January 2022
- Large Language Models are Zero-Shot Reasoners. Kojima et al., arXiv, May 2022
Topic: Prompt engineering
Last updated